Quantifying soil production rates is central to understanding landscape evolution, critical-zone development, and carbon cycling, but commonly relies on cosmogenic nuclide methods conditioned on the steady-state status of soil thickness. Here we introduce a two-sample test for steady-state that uses the ratio of <sup>10</sup>Be-inferred erosion rates from soil and saprolite. Across ridgetops spanning a wide range of erosion rates, land uses, and soil thicknesses, the test reveals both steady- and non-steady-state soils, sometimes on adjacent ridges. We numerically modelled soil disturbances to explore the implications of non-steady-state soils for <sup>10</sup>Be concentrations, erosion-rate estimates, and inferred soil production functions under a range of disturbance scenarios. The models show that non-steady soils can create apparent variability in soil production functions. Our two-sample test enables widespread verification of steady state assumptions, strengthening <sup>10</sup>Be-based interpretations of erosion, weathering, and soil development in upland landscapes.